I recently worked in an agile CRM project as the scrum master and was able to learn some good tips of using scrum for CRM development.
I am of the belief that the waterfall days are coming to an end, mostly due to its unsustainable release cadence, inefficiency when reacting to the 'unknown' and the large amount of waste that goes towards tight controls and project management as opposed to focusing effort on product quality and continuous business value generation. In the world of CRM implementations this becomes even more evident because of various factors such as the ability of the platform to quickly produce business value and adapt in an iterative manner, even when the final outcome is yet to be discovered. CRM can be configured very quickly which allows you to achieve valuable work (increment) in iterations (sprints) that can be as short as 1-2 weeks. The fact that you can have stakeholders provide early feedback continuously (sprint reviews) allows you to adapt very quickly to changing demands and gives your users a chance to get familiar with CRM early on. Continuous feedback leads to more valuable user feedback because users have something tangible to review instead of having to imagine what everything will look like. If you are reading this you probably already understand the value of agile and I don't need to convince you further. Now let's look at the typical challenges we find with agile frameworks like scrum and some ideas from my experience that can make your life easier.
1. Your customer needs to fund the project and to do so they need the cost, scope and schedule upfront. This has been the hardest challenge, specially in the consulting world as organizations are used to think that to manage risk means to transfer to someone else (you) the risk of changes in scope, cost and schedule. This is an antiquated mentality when it comes to complex software development because you cannot plan for everything in advance and expect that the requirements you document today is exactly what you are going to need. The result is typically that consulting companies add a large amount of contingency and charge you more for any change in your original requirements (change requests). The end result is that your customer ends up paying more and not getting the optimal solution. What is worse, is that they will not know that the solution is not really doing what they need until it is too late, too much has been wasted and the budget cannot be increased. This is by far the main reason I see waterfall projects fail. But what can you do about it if your customer's culture is waterfall and they want a fixed-everything upfront before they can "fund the project"? I think the magic word here is 'trust'. The best you can do (other than finding another customer that understands that we are not in the 90s) is to earn their trust. Certainly this does not happen overnight but if you are able to build this trust you might eventually be able to propose 'giving it a try' and educating your customer on modern software development practices. The good thing about agile is that you see the results very quickly and you can start by funding a single sprint to see how things go (the bad part is that the first sprint is always the hardest!). One important tip is to make sure your customer trusts you enough to not think that you are just trying to make more money with them, this will not go well and is very far from your objective. In reality your objective is to save them money and increase their probability of success, they need to see that intention clearly. Regarding funding, you can help them switch from the 'fund a project' mentality to the mentality of 'fund a team that is going to give you optimal business value for your project(s)'.
2. You are not sure how to manage the CRM development and test environments for an agile project. This is one where you might have to try a few things and see what works best with your development team. My suggestion is to start with something as simple as possible and empower your development team to decide if they want to make changes to this during sprint retrospectives. The simplest is to have one (yes, one!) shared environment (CRM organization/tenant) for development and test. I have blogged in the past here about how to manage multiple dev and test CRM environments and some good practices about that, but I invalidate my own post when it comes to agile. You want to reduce waste and not have to spend time migrating code, merging solutions and fixing environment synchronization issues. This is why you should at least try to have everyone working on the same environment. You will certainly need some rules such as one developer cannot block the environment for debugging while everybody else needs it; but this is why the daily scrum exists. Make sure you have a daily scrum and that it is actually a 15 min maximum daily scrum and not some status report or defect logging meeting, do not invite anybody outside of your development team to the scrum meeting, this time should be wisely used so developers can synchronize and coordinate activities to be the most efficient. Developers can always have their own sandbox environments but these are not development environments as you should not promote code from sandbox to development.
Regarding test environment I initially had a separate test environment but the development team quickly realized that it was causing too much overhead with very little value. Instead, try to get everyone to test during the sprint (add testing to your definition of done). A very hard thing to do is to bring acceptance testing as part of the sprint and not an ad-hoc activity after the sprint. Get your product owner engaged to do the necessary acceptance testing prior to sprint review, and even consider incorporating UAT as part of your sprint. This is not easy but will give you increased agility and faster release cycles. If you succeed bringing acceptance testing as part of your sprint then it is more clear why you would not need a separate test environment.
3. You don't know how long your sprints should be and how much PBIs you should forecast for a given sprint. I think this is one of the places where CRM has an advantage over custom development, you can do a lot in very little time. Therefore, I would suggest to start with shorter sprints (maybe 2 weeks is a good start). Never do a sprint over 4 weeks or you might end up just doing waterfall! At first you will most likely underestimate and not complete all the PBIs by the end of the first sprint. This is not ideal but is acceptable as long as you don't "extend" the sprint (sprints are time boxed, never extend the end date of a sprint, no matter what). The reality is that at first you might struggle with sprint planning, this is normal, try to forecast how much you can complete in your first sprint by keeping in mind your definition of done (make sure you have one!). So think about how much time it will take you not just to code/configure but also to design, analyze, unit test, document, etc.) each PBI. Additionally you need to work with the product owner to select the most urgent PBIs while keeping a balance for the skills required to deliver the sprint. For example, if you have a dev team of 6 but only 1 person knows SSRS, then you cannot realistically include only SSRS reports in the sprint, even if those reports are the most urgent PBIs. You need to make sure your team skill set is proportionate to the skill set required to deliver the sprint otherwise you might end up with 1-2 very busy individuals while others have a hard time finding how to contribute. Additionally when you forecast your sprint make sure you reserve a 10% of total sprint time to do backlog refinement (most teams fail to do this). There are multiple estimation techniques such as planning poker, however, when forecasting your sprint always ask your team at the end to do a 'gut check' (are we confident to be able to deliver these PBIs within next sprint?). I have found gut checks to be very useful and accurate, everyone must keep in mind the time spent in meetings, backlog refinement, sprint reviews and retrospectives, etc. because all these activities are also part of your sprint, not just the development activities. So in reality you spend only about 80% of your sprint capacity developing PBIs (note that development includes design, code, unit test, etc). One more thing about sprints, try to keep the same duration of sprints and the same start and end day of the week. Eliminate any time between sprints (yes this can be hard). Sprint n+1 should start the day after sprint n finishes, this will give you greater agility, as opposed to having "ventilation" time between sprints to do activities that should actually have been done during the sprint.
4. You are uncertain about how you can incorporate UAT and deployment into the scrum process. This is one of those places where reality sometimes crashes with scrum. If you can get your customer to perform UAT during the sprint and you make that work then you are probably a scrum super ninja that is not learning anything new from this post. So that's the ideal scenario, but we know for various reasons this does not always work. In this case you will need to do as much acceptance testing with the product owner during the sprint as possible and then release to UAT as part of your definition of done. The problem is that your team will probably get interruptions for UAT support as the UAT testers are usually in a completely different cycle as your sprint cycles. You will have to deal with this and account for this during sprint planning, but again the more you can test earlier the better because you will get less UAT defects and distractions. In this case you might need a separate environment for UAT and you would need to forecast in your sprint planning the activities for UAT release. Regarding deployment to production, I have often heard the misunderstanding that in scrum you are supposed to deploy to production at the end of every sprint, that's not correct. You can release to production as often as you want, from multiple times per sprint to once very number of sprints. However, at the end of each sprint, your work should be "potentially releasable" according to the scrum guide, so you shouldn't have broken critical functionality at the end of the sprint, even if you don't plan to release to production. It is a good idea to release as often as you can, try to incorporate the DevOps approach and bring development and operations together as much as possible, this will help you achieve a faster release cycle and a smooth transition to production. Of course this is not something that you will get right from one day to the next, DevOps requires multiple changes and great amount of practice and experience to become efficient (beyond the scope of this post).
Scrum is an empirical process, so at its core it has transparency, inspection and adaptation. Make sure that you pay attention to these 3 values during your various events, the key is not how to successfully setup a scrum project but more about your agility to inspect and adapt your process to produce business value faster and reducing waste, that is your success criteria! There are certainly many other challenges and suggestions, I might blog more about this in the future, but I would conclude with one famous quote from scrum that couldn't be more accurate: scrum is a framework that is easy to understand but very hard to master.